Download as pdf or txt
Download as pdf or txt
You are on page 1of 24

JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 117, F04024, doi:10.

1029/2012JF002539, 2012

Mapping gravel bed river bathymetry from space


C. J. Legleiter1 and B. T. Overstreet1
Received 2 July 2012; revised 1 October 2012; accepted 5 October 2012; published 21 November 2012.

[1] Understanding river form and behavior requires an efficient means of measuring
channel morphology. This study evaluated the potential to map the bathymetry of two
clear-flowing, shallow (<3 m deep) gravel bed rivers <60 m wide from 2 m-pixel
WorldView2 satellite images. Direct measurements of water column optical properties
were used to quantify constraints on depth retrieval. The smallest detectable change in
depth was 0.01–0.04 m and the maximum detectable depth was 5 m in green bands but
<2 m in the near-infrared; lower sensor radiometric resolution yields less precise
estimates over a smaller range. An algorithm for calibrating a band ratio X to field
measurements of depth d proved effective when applied to spectra extracted from
images (R2 = 0.822 and 0.594 for the larger and smaller stream, respectively) or
measured in the field (R2 = 0.769 and 0.452). This procedure also identified optimal
wavelength combinations, but different bands were selected for each site. Accuracy
assessment of bathymetric maps produced using various calibration approaches and
image types indicated that: 1) a linear d vs. X relation provided depth estimates nearly
as accurate as a quadratic formulation; 2) panchromatic and pan-sharpened
multispectral images with smaller 0.5 m pixels did not yield more reliable depth
estimates than the original images; and 3) depth retrieval was less reliable in pools due
to saturation of the radiance signal. This investigation thus demonstrated the feasibility,
as well as the limitations, of measuring the bathymetry of clear, shallow gravel bed
rivers from space.
Citation: Legleiter, C. J., and B. T. Overstreet (2012), Mapping gravel bed river bathymetry from space, J. Geophys. Res., 117,
F04024, doi:10.1029/2012JF002539.

1. Introduction detailed surveys. A more synoptic perspective on fluvial


systems will require a different approach, and remote sens-
[2] The form and behavior of gravel bed rivers reflect ing is increasingly viewed as a viable alternative [Marcus
complex interactions among morphology, flow, and sedi-
and Fonstad, 2008, 2010]. Previous studies have demon-
ment transport. Understanding connections between form
strated the feasibility of deriving various types of river
and process is thus a principal objective of fluvial geomor-
information from image data, ranging from measurements of
phology, but progress toward this goal is hindered by the
lateral channel migration from historical air photos [e.g.,
difficulty of collecting basic data on topography, flow con-
Micheli and Kirchner, 2002] to suspended sediment con-
ditions, and bed material properties. Moreover, the logistical
centrations inferred from Landsat scenes [Mertes et al.,
constraints associated with traditional field methods for
1993; Kilham et al., 2012]. The most mature application of
measuring these attributes often limit investigations to short,
remote sensing to rivers is retrieval of water depth from
isolated study reaches. Although recent advances in instru-
passive optical images, primarily multi- or hyperspectral data
mentation, such as total stations [Keim et al., 1999], real- acquired from aerial platforms [e.g., Winterbottom and
time kinematic global positioning systems (RTK GPS)
Gilvear, 1997; Marcus et al., 2003; Lejot et al., 2007;
[Brasington et al., 2000], and terrestrial laser scanning
Legleiter et al., 2009; Flener et al., 2012]. In this study, we
[Hodge et al., 2009], have enabled more efficient data col- build upon these results by using field measurements and
lection, most research continues to focus on scales ranging
satellite imagery to evaluate the potential for mapping the
from a few to several tens of channel widths, often with little
bathymetry of gravel bed rivers from space.
consideration of the broader watershed context for these
[3] Knowledge of water depth is valuable for a number of
1
different purposes. For example, depth is an important
Department of Geography, University of Wyoming, Laramie, parameter for any hydrologic study that involves computing
Wyoming, USA.
river discharge and/or routing flows. From a geomorphic
Corresponding author: C. J. Legleiter, Department of Geography, perspective, the depth, along with the water surface slope,
University of Wyoming, Dept. 3371, 1000 E. University Ave., Laramie, exerts a primary control on the boundary shear stress that in
WY 82071, USA. turn drives bed material transport. In an ecological context,
©2012. American Geophysical Union. All Rights Reserved. the thickness and optical properties of the water column
0148-0227/12/2012JF002539

F04024 1 of 24
F04024 LEGLEITER AND OVERSTREET: MAPPING RIVER BATHYMETRY FROM SPACE F04024

determine the amount of solar energy that propagates to the the accuracy of depth estimates produced from various types
streambed to fuel primary production by benthic algae. For of satellite imagery.
managers interested in monitoring stream condition and
change, depth is one of the principal quantities used to 2. Methods
assess habitat quality. In addition, when combined with
information on water surface elevations, depth measure- 2.1. Study Area
ments can be used to obtain topographic input data for [6] To evaluate the feasibility of mapping river bathyme-
numerical modeling of flow and sediment transport. Sim- try from space, we examined two gravel bed streams in the
ilarly, sequential observations of channel morphology can Rocky Mountains, USA: the Snake River in Grand Teton
be used to identify areas of erosion and deposition and National Park and Soda Butte Creek (SBC) in Yellow-
hence to infer bed material transfer and storage [e.g., stone National Park (Figure 1). Both watersheds are
Ashmore and Church, 1998; Ham and Church, 2000]. snowmelt-dominated and generally exhibit clear water
A capacity to map river bathymetry efficiently, reliably, conditions during late summer low flows. The two rivers
and over large extents would thus benefit the riverine feature wandering planforms with both sinuous, single-
sciences in many ways. thread and more complex multi-thread segments. In addition,
[4] Remote sensing could provide such capability. The both streams are highly dynamic, with extensive morpho-
potential of this approach has been confirmed first through logic change occurring during spring runoff in many years,
empirical case studies [e.g., Winterbottom and Gilvear, including 2011. The steep, glaciated valley of SBC is com-
1997; Marcus et al., 2003] and more recently by consider- posed of weak Eocene volcanic rock and provides an abun-
ing the underlying physics [Legleiter et al., 2004, 2009]. dant sediment supply that drives channel change [Meyer,
Passive optical remote sensing of river bathymetry involves 2001]. The Snake River is regulated at Jackson Lake, but
measuring the amount of solar radiation reflected from the sediment delivered from tributaries below the dam [Erwin
channel, which depends not only on depth but also the tex- et al., 2011], along with large woody debris, contribute to
ture of the water surface, the concentration and composition frequent morphologic adjustments. We selected these rivers
of sediment and organic materials within the water column, for study because the complexity and dynamism of these
and the reflectance of the streambed. Moreover, all of these channels imply that remote sensing might be useful, if not
quantities vary as a function of wavelength l. To gain necessary, for characterizing their form, behavior, and
insight as to how the processes that govern the interaction of evolution.
light and water both enable and limit the remote sensing of [7] For the Snake River we focused on a meander called
rivers, Legleiter et al. [2004] used a radiative transfer Rusty Bend, shown in Figure 1c and described in Table 1.
model to isolate the effects of depth d, suspended sediment The channel curves smoothly to the right through this reach
concentration, and bottom reflectance RB(l) on the and has a relatively simple morphology consisting of a gravel
upwelling spectral radiance L(l) recorded by a remote bar along the inner (right, or north) bank and a vertical to
detector. For conditions representative of a shallow, clear- steeply sloping cutbank on the outside of the bend where the
flowing gravel bed river, this analysis indicated that depth Snake River erodes into a high terrace. For the purposes of
was the primary control on L(l) and that taking the loga- this investigation, the bar-pool topography of Rusty Bend
rithm of the ratio of two specific bands yielded an image- was attractive because of the broad range of depths encom-
derived quantity X linearly related to d. Subsequently, we passed: very shallow over the bar and up to 2.77 m in the pool
argued on theoretical grounds that spectrally-based depth along the outer bank. Also, a variety of substrates with dif-
retrieval would be feasible when d is small, the substrate is ferent reflectance characteristics were present within this
highly reflective relative to the overlying water column, reach: clean gravel, bright green benthic algae, and blocks of
and attenuation of light is dominated by pure water lighter-colored bedrock (Figure 2b).
absorption rather than scattering by suspended sediment. [8] Soda Butte Creek has been the subject of previous
Radiative transfer modeling, field-based reflectance mea- efforts to map river bathymetry from aerial platforms
surements, and bathymetric maps derived from hyper- [Marcus et al., 2003; Legleiter et al., 2004, 2009; Legleiter,
spectral image data supported these conclusions and 2012a] and thus provides a convenient opportunity to assess
confirmed the utility of a simple band ratio for remote whether similar information might be derived from satellite
measurement of water depths in certain types of rivers images. In this study, we focused on the Footbridge Reach,
[Legleiter et al., 2009]. a site for which ground-based topographic surveys have
[5] Although previous studies have shown that depth been conducted each year since 2005 [Legleiter, 2012b].
information can be retrieved from aerial images, the feasi- The reach comprises a sweeping meander and large point
bility of mapping river bathymetry from space has not been bar, which is now being incised by a chute channel that has
assessed. In this study, we used field measurements and captured much of the stream’s flow and could soon cut off
multispectral image data to explore the possibility of mea- the bend. Because the discharge is divided among multiple
suring gravel bed river depths from a satellite platform. Our channels and the image was acquired under base flow con-
research objectives were to: (1) characterize the inherent ditions, depths were generally quite shallow (Table 1) but
optical properties of the water column by obtaining novel in deeper pools were present in the middle of the chute chan-
situ measurements of absorption and attenuation in a pair of nel, along the west bank above the bend apex, and against
clear-flowing gravel bed streams, (2) establish relationships the east bank at the lower end of the reach. The smaller size
between depth and reflectance based on field spectra from of SBC also implied that mixed pixels along channel mar-
these rivers, and (3) evaluate different approaches for cali- gins would be more extensive and potentially more prob-
brating image-derived quantities to flow depth and (4) assess lematic than for the Snake River.

2 of 24
F04024 LEGLEITER AND OVERSTREET: MAPPING RIVER BATHYMETRY FROM SPACE F04024

Figure 1. (a) Map indicating the location of study sites in Grand Teton National Park (GTNP) and Yel-
lowstone National Park (YNP), USA. WorldView2 images of the (b) Snake River and (d) Soda Butte
Creek are shown, with insets highlighting (c) Rusty Bend and (e) the Footbridge Reach.

2.2. Field Data Collection


[9] On the Snake River, we surveyed channel bed topog-
raphy using a high-precision (2–3 cm, both horizontal and Table 1. Channel Characteristics for Study Reaches
vertical) real-time kinematic (RTK) GPS receiver. Eleva- Snake River Soda Butte Creek
tions were measured at points arranged along cross-sections Rusty Bend Footbridge Reach
that traversed exposed bars and shallow areas of the wetted
Radius of curvature (m) 84 42
channel. Depths were determined by subtracting the bed Mean wetted width (m) 58 19
elevation for in-stream survey points from the elevation Channel bed slope 0.0035 0.0065
recorded at the water’s edge along each transect [Legleiter Mean da  std. dev. (m) 1.23  0.57 0.22  0.15
et al., 2011a]. For areas that were too deep to wade safely, Maximum d (m) 2.77 0.69
nb 3276 655
the GPS receiver was mounted on a cataraft and configured
to record water surface elevations while communicating with a
The symbol d denotes the pixel-scale mean depth determined from field
an echo sounder that measured flow depths. Measurements measurements via ordinary block kriging.
b
The symbol n is the number of d values for the reach.

3 of 24
F04024 LEGLEITER AND OVERSTREET: MAPPING RIVER BATHYMETRY FROM SPACE F04024

Figure 2. (a) Field measurements of flow depth and spectral reflectance were obtained from a cataraft;
the spectroradiometer extends out over the water from the rear of the cataraft (right in this photo).
(b) Blocks of light-colored bedrock present along Rusty Bend. (c) Underwater photograph of the scissor lift
apparatus used to measure downwelling radiant energy at different depths within the water column.

were obtained along a series of transects across our Rusty important purposes: 1) up-scaling observations collected at
Bend study reach as well as longitudinal profiles recorded as points to the dimensions of an image pixel [Bailly et al.,
the cataraft traveled downstream. Over 22 km of the Snake 2010]; and 2) producing continuous depth maps for com-
River was surveyed in this manner during a 10-day period in parison with the remotely sensed bathymetry. To account for
August and September 2011, resulting in a total of 73,686 the non-convex geometry of these meandering rivers, which
echo sounder-based depth measurements. In addition, an implies that conventional Euclidean distances are not a valid
acoustic Doppler current profiler (ADCP) deployed from a metric [e.g., Little et al., 1997], data were transformed from
kayak provided additional depth observations. To ensure the original Cartesian reference frame to an orthogonal cur-
that the depth measurements obtained via the three methods vilinear, channel-centered coordinate system defined by a
were consistent with one another, we compared echo streamwise axis s along the channel centerline and a cross-
sounder and ADCP readings to the closest wading depth. stream, or normal, n axis oriented perpendicular to the cen-
This analysis indicated that in shallow areas where the data terline [Smith and McLean, 1984; Legleiter and Kyriakidis,
sets overlapped, the echo sounder depths were biased shal- 2006]. The spatial structure of each reach was then quanti-
low by 7 cm and the ADCP depths biased deep by 3 cm fied using anisotropic variograms, as described by Legleiter
relative to the wading depths. The mean differences between [2012d]. Briefly, variograms summarize the degree to
the echo sounder and ADCP depths and the corresponding which observations are spatially correlated with one another
wading points were used to adjust the echo sounder and as a function of the distance and direction (i.e., lag vector)
ADCP data to match the wading depths. The resulting, between pairs of points and thus provide information on
combined bathymetric field data set was then used for cali- overall variability as well as the length scales over which this
bration and validation of image-derived depth estimates. variability is expressed along and across the channel. In this
[10] For the smaller Soda Butte Creek, all field measure- study, de-trending was not necessary because we used depth
ments were obtained by wading, which allowed us to access measurements rather than bed elevations that would have
all but the deepest portions of the channel. Field data col- been influenced by the overall slope of the channel. Sample
lection at the Footbridge Reach involved using the RTK GPS variograms were calculated for the s and n directions by
and a robotic total station to complete a detailed, terrain- restricting the angular tolerances associated with each lag
sensitive survey that emphasized important breaks in slope vector class. Variogram model parameters were estimated
such as the top and base of banks [e.g., Lane et al., 1994; first manually using an iterative graphical procedure and then
Wheaton et al., 2010]. In addition, water surface elevations refined using a weighted least squares algorithm that
were measured along channel margins and used to calculate emphasized shorter lag distances with larger numbers of pairs
depths for in-stream points as the difference between water [Zhang et al., 1995; Pardo-Iguzquiza, 1999].
surface and bed elevations. An earlier study confirmed that [12] Although field measurements were collected at dis-
depths determined in this manner agreed closely with crete points, depth estimates derived from satellite images
depths measured directly with a ruler, which would have pertain to larger areas of space, represented by image pixels.
been a less efficient means of obtaining these data [Legleiter, To account for this difference in scale, or change of support
2012c]. [e.g., Atkinson and Curran, 1995], we employed a geosta-
tistical algorithm known as ordinary block kriging (OBK),
2.3. Geostatistical Analysis described in general by Goovaerts [1997] and in a remote
[11] Before using the depth measurements described sensing context by Bailly et al. [2010]. Here, we briefly
above to calibrate image-derived depth estimates and assess summarize the rationale for and implementation of this
their accuracy, the original field data were processed procedure. Essentially, we used OBK to upscale point
using geostatistical techniques. These analyses served two observations to the dimensions of an image pixel and obtain

4 of 24
F04024 LEGLEITER AND OVERSTREET: MAPPING RIVER BATHYMETRY FROM SPACE F04024

spatially distributed estimates of the pixel-scale mean depth. streams. Attenuation of light was characterized by measur-
Also, because some pixels contained multiple points, this ing the total amount of incident solar radiation, referred to as
analysis served to avoid the redundancy that would have the downwelling spectral irradiance Ed(l), at various depths
occurred if the original field measurements were paired within the water column. These irradiance profiles were
directly with specific image pixels. Instead, the OBK algo- collected by connecting the ASD to an upward-facing cosine
rithm yielded a single estimate of the average depth within response detector that integrated radiant energy arriving
each pixel, regardless of the number of points present within from all directions within the upper hemisphere to obtain
that area. To compute OBK estimates, we first created a fine- Ed(l); the fore-optic was attached to a waterproof cable and
scale grid of discretization points for each reach, consisting mounted on an adjustable scissor lift to position the sensor at
of 16 points per pixel, and transformed these prediction different depths (Figure 2c). These measurements were used
locations to the same channel-centered coordinate system as to calculate the diffuse attenuation coefficient Kd(l), an
the field data. A depth estimate for each discretization point apparent optical property that quantifies the rate at which
was then calculated using an ordinary kriging algorithm that light is attenuated with distance traveled through the water
incorporated the anisotropic variogram models described column, following the procedure outlined by Mishra et al.
above [Legleiter and Kyriakidis, 2008]. Taking the average [2005] and applied to the Platte River by Legleiter et al.
of the ordinary kriging estimates for the 16 points within [2011b]. In addition, we used a WET Labs ac-9 to directly
each pixel yielded the block kriging estimate of the pixel- measure two inherent optical properties of the water column,
scale mean depth [Goovaerts, 1997, p. 152]. For purposes of the absorption and attenuation coefficients, a and c. These
calibration and validation, OBK depth estimates were com- optical data were collected on several dates along the Snake
puted only for those pixels containing one or more point River and SBC. Ancillary data in support of these mea-
measurements. In addition, to examine spatial patterns of surements included water samples analyzed for suspended
depth retrieval accuracy, continuous field-based bathymetric sediment concentration and turbidity readings made with a
maps were generated by estimating via OBK the mean depth Eureka Environmental Manta2 multiprobe.
for all pixels within each reach. [16] Measuring water column optical properties allowed
us to examine two important constraints on remote sensing
2.4. Spectral Characteristics of Gravel Bed Rivers: of river bathymetry: the precision of spectrally-based depth
Field Measurements and Data Analysis estimates and the maximum depth detectable by an imaging
[13] One of our long-term research objectives is to build a system. This analysis was based on the early work of Philpot
more thorough database on the spectral characteristics of [1989], which was revisited in a fluvial context by Legleiter
fluvial environments, where only a few quantitative obser- and Roberts [2009]. The original publications provide
vations of water column optical properties and bottom additional detail, but only the key results relevant to this
reflectance have been made [Legleiter et al., 2009, 2011b]. investigation are highlighted herein. Values of Kd(l) deter-
In this study, we began the process of compiling a spectral mined from our field measurements were used to calculate
library for rivers by collecting field spectra and measuring bathymetric precision as
the apparent and inherent optical properties of the water  
1 DLN ðlÞ
column in a pair of clear-flowing gravel bed streams. Dd ðlÞ ¼ ln 1  expf2Kd ðlÞd0 g ð1Þ
[14] Reflectance spectra were recorded from above the 2Kd ðlÞ L B ðl Þ
water surface using an Analytical Spectral Devices (ASD) where Dd(l) is the smallest detectable difference in depth at
FieldSpec3 spectroradiometer that measured wavelengths an initial depth d0, DLN(l) is the sensor’s noise-equivalent
from 350–1025 nm with a 1 nm sampling interval; only the delta radiance (essentially the smallest change in brightness
400–850 nm region was considered here due to noise at both the system can resolve), and LB(l) is that portion of the total
ends of the spectrum. A 100% reflectant Spectralon cali- radiance signal that has interacted with the bottom and is
bration panel was used to establish a white reference prior to thus related to depth. Because DLN(l) and LB(l) are both
each round of measurements. For data collection along the spectral radiance values, the units cancel and only the ratio
Snake River, the ASD was mounted on a rod extending from DLN(l)/LB(l) is significant in equation (1). This ratio serves
the rear of the cataraft and configured to record spectra once as a convenient index of the detectability of the bottom for a
each second as we traversed a series of transects across given river and sensor configuration. In this study, we cal-
Rusty Bend (Figure 2a). Flow depths were recorded simul- culated Dd(l) values by specifying DLN(l)/LB(l) = 0.01,
taneously using the survey instrumentation described above, based on prior radiative transfer modeling [Legleiter and
providing the paired observations of depth and reflectance Roberts, 2009], and using 2Kd(l) as an effective attenua-
needed to develop bathymetric mapping algorithms. More- tion coefficient, following Philpot [1989] and Maritorena
over, these data extended the range of river conditions under et al. [1994]. Similarly, the maximum detectable depth
which spectra have been measured from shallow, wadeable occurs when the difference between the at-sensor radiance
streams [Legleiter et al., 2009] to a deeper, larger channel and the radiance from a hypothetical infinitely deep water
with diverse bottom types (Figure 2b). For SBC, field column is equivalent to the sensor’s DLN(l) and was calcu-
spectra were recorded at points accessed by wading and lated as
depths measured with a ruler. Additional detail on the  
acquisition and processing of reflectance data were provided 1 DLN ðlÞ
dmax ðlÞ ¼ ln : ð2Þ
by Legleiter et al. [2011b]. 2Kd ðlÞ LB ðlÞ
[15] Because water column optical properties influence the
feasibility of inferring depth from image data, we directly In this case, we used DLN(l)/LB(l) values of 0.1, 0.01,
measured several attributes of the water within our study 0.001, and 0.0001 to illustrate the effects of a greater bottom

5 of 24
F04024 LEGLEITER AND OVERSTREET: MAPPING RIVER BATHYMETRY FROM SPACE F04024

contrast between the substrate and water column and/or a agreement between field and image data, with depth points
more sensitive detector, either or both of which would cor- located within the wetted channel on the transformed
respond to smaller values of DLN(l)/LB(l). image.
[20] An important pre-processing step was the definition
2.5. Image Data Acquisition and Processing of in-stream masks for each image. These masks served to
[17] To evaluate the feasibility of mapping gravel bed isolate active channels and thus highlight variations in
river bathymetry from space, this study used satellite images reflectance within the water portion of each scene. In this
acquired by the WorldView-2 (WV2) sensor. This imaging study, binary, water-only masks were produced by display-
system became operational in January 2010 and features a ing the longest wavelength NIR band as a gray scale image,
unique combination of high spatial resolution (pixel sizes of inspecting the histogram of pixel values, and interactively
0.5 and 2 m for panchromatic and multiband images, adjusting the contrast stretch to determine a NIR reflectance
respectively) and multispectral measurement capabilities, threshold that effectively distinguished dark water from
with eight bands spanning visible and near-infrared (NIR) brighter terrestrial features. Pixels with NIR reflectance
wavelengths. In addition to the standard blue, green, red, and values below this threshold were included in an initial water
NIR bands, WV2 also includes coastal (400–450 nm), yel- mask that was refined using image processing operations:
low (585–625 nm), and red edge (705–745 nm) bands morphological opening to remove isolated pixels, interactive
potentially useful for depth retrieval from shallow streams as segmentation to select in-stream image objects, and mor-
well as a longer-wavelength NIR band (860–1040 nm) that phological closing to fuse small gaps [Legleiter et al.,
could be used to discriminate land from water. This satellite 2011a]. The resulting raster masks then were converted to
also features advanced pointing technology that allows for vector representations that enabled manual editing to remove
off-nadir viewing, reduced revisit times, and precise geo- persistent shadows along steep cut banks, for example. The
metric positioning, with nominal geo-referencing accuracies vector masks were rasterized and applied to the original
on the order of 4 m (Digital Globe, data available at http:// images. Finally, the resulting in-stream images were spa-
worldview2.digitalglobe.com/about/, 2012). tially filtered using a 3  3 pixel Wiener smoothing filter
[18] WV2 images of the Snake River and SBC were that has been shown to improve depth retrieval performance
acquired on 13 September 2011 (Figure 1). Deliverables [Legleiter, 2012c].
included geo-referenced multispectral and panchromatic
data sets and supporting metadata. In addition to the original 2.6. Spectrally-Based Depth Retrieval
images, which were not radiometrically calibrated and [21] Mapping river bathymetry from satellite images
consisted of raw digital numbers, we received atmospherically- requires a quantitative relationship between water depth d
corrected data processed to units of apparent surface reflec- and reflectance R(l) in one or more spectral bands; radiance
tance in-house by DigitalGlobe. Depth retrieval performance or raw digital numbers also could be used for this purpose.
was evaluated for the multispectral reflectance images as Efforts to establish such relationships are complicated,
well as the higher spatial resolution panchromatic data sets. however, by the influence exerted on the remotely sensed
In addition, we considered hybrid, pan-sharpened images that signal by several other factors, including variations in bottom
consisted of eight spectral bands, like the original multi- reflectance, water column optical characteristics, water sur-
spectral images, but had a smaller 0.5 m pixel size equivalent face roughness, and atmospheric conditions. The availability
to the panchromatic images; these images were generated of multiple spectral bands provides some leverage for iso-
using the Gram-Schmidt spectral sharpening tool in the lating the effect of depth, and Legleiter et al. [2009] showed
ENVI software. The three different types of images (multi- that under appropriate circumstances and for certain combi-
spectral, panchromatic, and pan-sharpened) allowed us to nations of wavelengths, the image-derived quantity
assess the relative significance of spatial and spectral reso-
lution for remote sensing of river depths.  
Rðl1 Þ
[19] Geo-referencing of the WorldView-2 image for the X ¼ ln ð3Þ
Snake River was highly accurate, with stream banks and Rðl2 Þ
other distinctive features in the image closely aligned with
our field maps; no further geometric correction of this data is linearly related to depth. The physical basis for ratio-based
set was needed. For SBC, however, the spatial referencing of depth retrieval was examined in detail by Legleiter et al.
the original image was not as reliable, with many of the [2004, 2009] and is only summarized herein. The upwelling
points at which depths were measured plotting outside spectral radiance from a clear-flowing, shallow stream
the wetted channel when overlain on the image. Given the channel is primarily a function of depth and bottom albedo.
smaller size of this stream and the abrupt variations in Whereas the reflectances of various substrates tend to be
depth over the scale of a single, 2 m image pixel, improved within a few percent of one another at a given wavelength
geo-referencing was required. The necessary refinement and thus have similar band ratio values, the absorption
was achieved by digitizing a wetted channel polygon on coefficient of pure water increases by an order of magnitude
the image and comparing this feature to a polygon created from the blue into the NIR portion of the spectrum. As a
from water surface elevation points surveyed in the field. result, the reflectance in the longer wavelength band with
The parameters of an affine transformation were then stronger absorption, l2, decreases more rapidly as depth
iteratively adjusted so as to maximize the area of overlap increases than does R(l1) and the ratio X increases with depth
between the image- and field-based channel polygons and while remaining relatively insensitive to differences in bot-
the original image transformed using the optimal para- tom type. Taking the logarithm of the band ratio accounts for
meters [Legleiter, 2012a]. This algorithm greatly improved the exponential attenuation of light by the water column.

6 of 24
F04024 LEGLEITER AND OVERSTREET: MAPPING RIVER BATHYMETRY FROM SPACE F04024

Figure 3. (a) Field spectra recorded at Rusty Bend of the Snake River; the median and interquartile range
of n = 904 samples are plotted. (b) Field spectra convolved to the spectral band passes of the WorldView-2
(WV2) sensor. (c) Image spectra from the WV2 image of Rusty Bend, extracted from the unique pixels at
which flow depths were measured in the field (n = 1638). (d) Image spectra corrected to better match the
convolved field spectra by adding the mean difference between the field and image spectra to the original
image spectra.

[22] This approach to retrieving water depth from spectra for each band and then added this correction factor to
remotely sensed data requires: 1) selecting a suitable pair of the image spectra to force a closer agreement between the
wavelengths; and 2) calibrating a linear relation between d field measurements and remotely sensed data (Figure 3d).
and X. Both of these objectives can be achieved by Optimal OBRA was then performed for the convolved field spectra
Band Ratio Analysis, or OBRA [Legleiter et al., 2009]. and the resulting d vs. X relation applied to both the original
Given paired observations of depth and reflectance, this image and the image corrected to better match the field
algorithm performs regressions of d on X for all possible spectra. This analysis thus allowed us to evaluate the per-
combinations of numerator (l1) and denominator (l2) formance of bathymetric mapping algorithms calibrated via
wavelengths and identifies the optimal band ratio as that field spectroscopy.
which yields the highest coefficient of determination R2; the [25] Although OBRA exploits the spectral information
corresponding regression equation provides a calibrated d available from multiband data sets, this procedure is not
vs. X relation. Because depth is regressed against X values applicable to panchromatic images that achieve a higher
defined by all possible band combinations, OBRA also is spatial resolution by integrating electromagnetic energy
useful for examining spectral variations in the nature and from across the spectrum. To assess whether this additional
strength of the relationship between d and X, which can be spatial detail might enable reliable depth retrieval without
visualized by plotting the matrix of R2(l1,l2) values as a requiring spectral information, we evaluated the bathymetric
matrix. mapping capabilities of the 0.5 m-pixel panchromatic WV2
[23] In this study, we performed OBRA of both field images of the Snake River and SBC. Rather than relating d
spectra collected along the Snake River and SBC and image to X as for the multispectral data, we used the linear trans-
spectra extracted from WV2 images of these streams. form, or Lyzenga [1981] algorithm, a more traditional
Analysis of the field spectra made use of the collocated approach to depth retrieval in rivers [e.g., Winterbottom and
depths measured with the echo sounder or by wading, Gilvear, 1997; Fonstad and Marcus, 2005; Flener et al.,
whereas OBRA of image spectra used pixel-scale mean 2012]. In this case, the image-derived quantity related to
depths estimated via OBK. For the larger and deeper Snake depth is given by
River, we also performed a modified version of OBRA that
included not only a linear X term but also an X2 term in each Y ¼ ln½R  R∞  ≈ ln½DN  minðDN Þ þ 1 ð4Þ
regression, based on the finding of Dierssen et al. [2003]
that a quadratic equation improved bathymetric retrievals where R represents the reflectance integrated over the WV2
in areas of greater depth. sensor’s panchromatic band and R∞ is the reflectance from a
[24] To assess whether d vs. X relations derived from field hypothetical, infinitely deep water column that encompasses
spectra could be used to infer depth from remotely sensed the contributions from the water column, water surface, and
data, we convolved the original field spectra, which were atmosphere but which lacks any signal from the bottom; the
recorded with a sampling interval of 1 nm (Figure 3a), to the latter term is thus known as the deep-water correction.
specific band passes of the WV2 sensor. The convolved Computationally, we implemented this algorithm using the
spectra shown in Figure 3b thus had a similar, though less raw digital numbers (DN) from each gray scale image and
well-resolved, shape as the original measurements, but the used the minimum pixel value from the in-stream portion of
convolved field spectra differed from the image spectra in the scene as an estimate of R∞; the last term on the right is
absolute magnitude due to residual atmospheric effects added to avoid taking the logarithm of zero. Pixel-scale
(Figure 3c). To account for this difference, we subtracted the mean depths estimated from the original field measurements
mean reflectance of the image spectra from that of the field via OBK were then regressed against Y values for the

7 of 24
F04024 LEGLEITER AND OVERSTREET: MAPPING RIVER BATHYMETRY FROM SPACE F04024

Figure 4. Diffuse attenuation coefficient spectra Kd(l) calculated from field measurements of the down-
welling spectral irradiance Ed(l) at various depths within the water column. For each irradiance profile
Ed(l) was measured at 10–12 different depths from just beneath the water surface to 0.6 m. Data were col-
lected from the clear-flowing Snake River (SR) and Soda Butte Creek (SBC) on the indicated dates in
2011, and from the more turbid Platte River (PR) in Nebraska in 2010 [Legleiter et al., 2011b].

corresponding locations to establish d vs. Y calibration water column by measuring vertical profiles of downwelling
relations. spectral irradiance and calculating values of the diffuse
[26] Calibration relationships obtained via OBRA or the attenuation coefficient Kd(l) (Figure 4). Data sets from six
linear transform were based on randomly selected subsets different dates resulted in similar Kd(l) spectra that varied
(50%) of the pixel-scale mean depths derived from the little with wavelength through the visible but increased
original field measurements. The remaining d values were abruptly at 700 nm due to a sharp rise in the absorption
reserved and used to validate image-derived depth estimates; coefficient of pure water, denoted by aw(l), in the NIR.
accuracy assessment involved calculating mean errors (an Similarly, the dip in Kd(l) around 810 nm is associated with
indication of bias) and root mean square errors (RMSE), as a decrease in aw(l) at this wavelength. Because these
described by Legleiter et al. [2011a], and examining residual streams had only small amounts of suspended sediment or
maps. In addition, we performed regressions of observed dissolved organic matter, their optical properties were dic-
(pixel-scale mean depths obtained via OBK) versus pre- tated primarily by those of pure water. To illustrate the
dicted (derived from the image) depths [Pineiro et al., 2008]. contrast between these clear-flowing gravel bed rivers and a
We conducted this type of analysis for Rusty Bend and the more turbid sand-bed channel, a Kd(l) spectrum from the
Footbridge Reach and considered several different calibra- Platte River was added to Figure 4 [Legleiter et al., 2011b].
tion approaches and image data types: 1) linear vs. quadratic The higher turbidity (49.5 NTU) and suspended sediment
OBRA for the Snake River; 2) OBRA of convolved field concentration (161 mg/L) of the Platte River lead to Kd(l)
spectra, with or without a correction applied to the image to values 2–5 times greater than those observed in our study
force better agreement with the field spectra; and 3) pan- area throughout the visible, with the greatest difference
chromatic vs. pan-sharpened WV2 images. occurring in shorter blue-green wavelengths more suscepti-
ble to scattering by suspended sediment. In the NIR, where
3. Results optical properties were driven primarily by pure water
absorption, the Kd(l) spectra for the three rivers converged
3.1. Optical Characteristics of Gravel Bed Rivers but remained higher for the Platte.
3.1.1. Water Column Optical Properties [28] The diffuse attenuation coefficient is an apparent
[27] Ancillary data collected along the Snake River and optical property influenced by variations in the ambient light
SBC confirmed our visual impression of exceptional water field, in addition to the spectral characteristics of the water
clarity. Turbidity values were consistently low (2–3 Neph- itself. To isolate the effects of the water on the interaction of
elometric Turbidity Units, or NTU) and suspended sediment light with the river, we made direct measurements of inher-
concentrations were minimal: 2 mg/L for each of three water ent optical properties with an ac-9. In situ observations of the
samples from the Snake River and one sample from SBC. attenuation a(l) and absorption c(l) coefficients quantita-
We characterized the interaction of solar energy with the tively verified the clarity of both streams (Figure 5). Again,

8 of 24
F04024 LEGLEITER AND OVERSTREET: MAPPING RIVER BATHYMETRY FROM SPACE F04024

Figure 5. Inherent optical properties of the water column measured with an ac-9 on the indicated dates
on the Snake River (SR) and Soda Butte Creek (SBC). The blue lines represent the attenuation coefficient
c and the black lines the absorption coefficient a. Also included are the absorption coefficients of pure
water and suspended sediment with a concentration of 2 g/m3, based on data included with the HydroLight
radiative transfer model [Mobley and Sundman, 2001].

the data sets agreed closely with one another, with higher the streambed LB(l), and the sensor’s noise-equivalent delta
values of a(l) and c(l) in the shortest- and longest- radiance DLN(l) [Philpot, 1989; Legleiter et al., 2004;
wavelength bands and the weakest absorption and attenua- Legleiter and Roberts, 2009]. To explore the implications of
tion in the green at 555 nm, implying that penetration of this important concept, we inserted observed values of Kd(l)
solar energy through the river was most efficient at this into equation (1) and calculated the smallest detectable
wavelength. Also included in Figure 5 are the absorption change in depth Dd(l) at a range of initial depths d0; the
coefficients of pure water and suspended sediment, based on ratio DLN(l)/LB(l) was held constant at 0.01 [e.g., Philpot,
a specified concentration of 2 mg/L (= 2 g/m3) and an optical 1989]. The results of this analysis are illustrated in
cross-section included with the HydroLight radiative transfer Figure 6a, based on an irradiance profile from the Snake
model [Mobley and Sundman, 2001]. These data are con- River; Dd(l) values for other sites and dates were nearly
sistent with our measurements of a(l) for l > 600 nm, where identical because Kd(l) values for the various data sets were
sediment has little influence on overall water column so similar. For this example, at a wavelength of 700 nm a
absorption. For shorter wavelengths more strongly absorbed difference in depth of 0.01 m or less would be detectable at
by suspended mineral matter, the sum of the pure water depths up to 0.4 m, and even at a depth of 1 m Dd(l)
and suspended sediment absorption coefficients agrees well remained less than 0.04 m. These calculations imply that if
with our field data, implying that a simple two-component the imaging system is sufficiently sensitive and the bottom is
optical model might be a sufficient description of these well-illuminated and highly reflective, precise depth esti-
streams. The difference between the absorption and attenu- mates could be derived from remotely sensed data. Note,
ation coefficients is due to the effects of scattering, primarily however, that less sensitive instrumentation (i.e., larger
by suspended sediment [Legleiter et al., 2011b]. Overall, DLN(l)), less reflective substrates, and/or smaller amounts
the relatively simple optical characteristics and great clarity of incident radiation (i.e., smaller LB(l)), would lead to
of the Snake River and SBC implied that these channels larger Dd(l) and less precise depth estimates. For example,
would be amenable to spectrally-based depth retrieval. images acquired at greater solar zenith angles (e.g., early or
[29] Our measurements of water column optical properties late in the day, or during the spring or fall) and/or encom-
also allowed us to quantify some of the limitations associ- passing darker streambed materials (e.g., basalt) will have
ated with remote sensing of river bathymetry. Because smaller values of LB(l) and hence larger values of Dd(l) for
imaging systems have a finite capacity to detect small a given sensor.
changes in the amount of upwelling spectral radiance, truly [30] The finite sensitivity of imaging systems also dictates
continuous depth maps cannot be derived from digital image that river bathymetry can only be mapped up to some max-
data. Instead, the smallest change in depth a particular sensor imum detectable depth. Again, dmax(l) depends on the
can resolve depends on the rate at which light is attenuated optical properties of the water column, summarized in terms
by the water column, the amount of radiance reflected from of an effective attenuation coefficient 2Kd(l), and the ratio

9 of 24
F04024 LEGLEITER AND OVERSTREET: MAPPING RIVER BATHYMETRY FROM SPACE F04024

Figure 6. Example calculations of the precision of image-derived depth estimates Dd(l) and maximum
detectable depth dmax(l) based on Kd(l) values from the Snake River collected on 9 September 2011
(Figure 4). (a) The Dd values calculated for a wavelength of l = 700 nm for a range of initial depths
d0 and a DLN/LB ratio of 0.01 using equation (1). (b) The dmax(l) spectra calculated for the DLN(l)/
LB(l) values indicated in the legend using equation (2).

DLN(l)/LB(l), where in this case DLN(l), the smallest dif- mapped remotely with a high degree of precision across the
ference in radiance the imaging system can resolve, is set range of depths, typically less than 2.5 m, observed in many
equal to the difference between the at-sensor radiance and gravel bed rivers.
the radiance from optically deep water [Philpot, 1989; 3.1.2. Field Spectroscopy and Relationships Between
Legleiter et al., 2004]. Maximum detectable depths were Reflectance and Water Depth
calculated via equation (2) for a range of DLN(l)/LB(l) [31] A primary objective of this study was to establish
values; the results of this analysis, based on a Kd(l) spec- quantitative relationships between reflectance and water
trum from the Snake River, are summarized in Figure 6b. depth for the two streams we examined. Field spectra mea-
For DLN(l)/LB(l) = 0.0001 (i.e., a highly sensitive instru- sured from above the water surface on the Snake River and
ment), dmax(l) was 10.9 m in the green wavelengths where SBC were processed following Legleiter et al. [2011b]. Each
water column attenuation was weakest. As pure water set of field spectra, together with the corresponding depth
absorption increased through the red and NIR, dmax(l) measurements, was then used as input to the OBRA algo-
decreased to less than 2 m for l > 730 nm. For a system with rithm described in Section 2.6. This procedure identified
a lower radiometric resolution, corresponding to a two-order combinations of wavelengths that were sensitive to varia-
of magnitude decrease in DLN(l)/LB(l) to 0.01 (the value tions in depth but robust to other factors that might influence
used to calculate the Dd(l) values in Figure 6a), depths up reflectance, such as substrate heterogeneity or sun glint from
to 5.45 m could be detected at 560 nm, but dmax(l) = 0.67 in the water surface. The results of this analysis are summa-
the NIR at 760 nm. These results implied that a sensor rized in Figure 7 using OBRA matrices that represent the
capable of resolving small changes in radiance would be strength of the linear relation between d and X for all pos-
crucial to mapping bathymetry across a broad range of sible band combinations. Figure 7a indicates that for Rusty
depths. Similarly, multispectral data would allow for selec- Bend, where depths reached up to 3 m, a strong (R2 = 0.887)
tion of bands well-suited for depth retrieval across this relation between d and X was obtained for a green numerator
range. The maximum detectable depth also would be influ- band and red denominator band. Moreover, the OBRA
enced by the nature of the fluvial environment itself. For a matrix shows that a broader range of wavelengths would
given imaging system (i.e., a fixed DLN(l)), dmax(l) have resulted in d vs. X relations nearly as strong: numerator
depends on the bottom contrast between the substrate and bands less than 550 nm resulted in R2 > 0.8 for
water column, implying that highly reflective substrates and/ 575 < l2 < 720 nm. For this relatively deep reach, the NIR
or clear water would favor depth retrieval from deeper portion of the spectrum was not useful because strong
channels. Also note that the absolute magnitude of LB(l) is absorption by pure water lead to saturation of the reflectance
significant, such that images acquired under less well-illu- signal in pools. Similarly, the decrease in predictive power at
minated conditions (i.e., early or late in the day, or at high l2 = 675 nm was due to chlorophyll absorption by benthic
latitudes) would result in smaller values of dmax(l). In any algae present on the streambed; the reduced bottom albedo
case, our field measurements of water column optical prop- resulted in a smaller reflectance and a somewhat weaker
erties, together with the analytical framework represented by relation between d and X in this band.
equations (1) and (2), implied that bathymetry could be

10 of 24
F04024 LEGLEITER AND OVERSTREET: MAPPING RIVER BATHYMETRY FROM SPACE F04024

Figure 7. Optimal band ratio analysis (OBRA) of field spectra from (a) Rusty Bend and (b) the Foot-
bridge Reach. The color scale represents the coefficient of determination (R2) value for regressions of d
on X, where X is an image-derived quantity defined via equation (3), for all possible band combinations.
The numerator, l1, and denominator, l2, wavelengths defining the optimal band ratio are listed in the inset
of each panel, along with the corresponding regression equation, R2 value, and standard error.

[32] Results from SBC were similar, but the optimal band were small, however, on the order of 0.2–0.4 m, and this
ratio produced an even higher regression R2 (0.975) for a cross-section illustrated the robust performance of the
pair of longer wavelengths, and a number of other bands OBRA relation across a range of depths up to 2.75 m. This
would have yielded d vs. X relations nearly as strong level of accuracy was noteworthy due to the pronounced
(Figure 7b). In contrast to Rusty Bend, NIR wavelengths differences in bottom reflectance along this transect
were effective denominator bands for any numerator (Figure 2c). Whereas most of the cross-section consisted of a
l1 < 730 nm. In this case, the NIR was more useful due to gravel substrate with some degree of algal coating, several
shallower depths along the Footbridge Reach. Strong large blocks of clay bedrock were located near the outer
absorption by pure water at these wavelengths implied that bank. This cohesive material was noticeably lighter-colored
small changes in depth would produce in large changes in than the surrounding gravel and resulted in large spikes in
reflectance, resulting in very strong d versus X relations. reflectance in this part of the channel, as shown in Figure 8
Because depths were so shallow, the saturation that limited for the OBRA denominator band. The OBRA-based depth
the utility of the NIR on the Snake River was less of an issue estimates were shallower at these locations but not to the
on SBC. In general, stronger d vs. X relations at longer NIR degree that the abrupt increases in reflectance might seem to
wavelengths could be expected for shallower streams. dictate. These results thus confirmed that OBRA enabled
Low concentrations of suspended sediment and dissolved effective depth retrieval in the presence of highly heteroge-
organic matter and highly reflective substrates also favor neous substrates, cited by Legleiter and Roberts [2005] as
remote bathymetric mapping [Legleiter et al., 2009]. Both one of the primary advantages of this technique. This study
rivers examined in this study satisfied these criteria, and supported this conclusion in the context of a deeper river
analysis of field spectra showed that spectrally-based with more variable bottom reflectance than had been con-
depth retrieval would not only be feasible but potentially sidered previously.
highly accurate. [34] These results, though encouraging, were derived from
[33] An example of this capability is given in Figure 8, field spectra that provided essentially continuous reflectance
which shows data collected on a transect across Rusty Bend. data with a 1 nm sampling interval. To assess whether strong
Depths measured by the echo sounder agreed closely with relations between d and X could be derived from image data
depths calculated from the field spectra using the OBRA that integrate reflectance over a smaller number of broader
relation from Figure 7a. Correspondence between the bands, we convolved the original field spectra to match the
observed and predicted profiles was best over the shallow WV2 sensor response function, as described in Section 2.6
bar surface on the right side of the channel (Figure 8, left). and illustrated in Figures 3a and 3b. The convolved spectra
Through the middle of the stream, the OBRA relation consisted of eight discrete bands and were subjected to
resulted in slight over-predictions of depth, and the spec- OBRA to quantify the extent to which reduced spectral
trally-based estimates were shallower than the echo sounder resolution might compromise the ability to retrieve
data in the pool along the outer bank. These discrepancies bathymetry from satellite images. The results of this analysis

11 of 24
F04024 LEGLEITER AND OVERSTREET: MAPPING RIVER BATHYMETRY FROM SPACE F04024

Figure 8. Cross-section of observed depths and image-derived estimates from a transect of Rusty Bend
are plotted on the right axes. On the left axes is the reflectance at 607 nm, the denominator wavelength for
the optimal band ratio for this reach, recorded as the cataraft traversed the cross-section.

are summarized in Figure 9a, which indicates only a slight Snake River posed a challenging test of this approach
decrease in the OBRA R2 to 0.839. For the convolved because flow depths ranged up to 3 m, far greater than the
spectra, the optimal WV2 band combination was a green shallow streams that have been the subject of our prior
numerator and yellow denominator, essentially the same remote sensing investigations [Legleiter et al., 2009, 2011a].
wavelengths as for the original field spectra. The OBRA In larger gravel bed rivers, saturation of the radiance signal
matrix also indicated that the blue band would have been an might lead to a non-linear relationship between d and X and
effective numerator with either the yellow or red band as a could preclude depth retrieval from pools. To account for
denominator. These results implied that the sensor’s broader this possibility we performed both linear OBRA, with a
bands contained sufficient spectral information for inferring single X term in the regression against field measurements of
depth and provided further evidence of the feasibility of d, and a quadratic OBRA in which an X2 term also was
mapping river bathymetry from space. included in the regression, based on a similar study by
Dierssen et al. [2003]. The resulting OBRA matrices are
3.2. Mapping River Bathymetry From Satellite presented in Figures 9b and 9c, which indicate that the same
Image Data band combination, a blue numerator with a yellow denomi-
[35] Our in situ measurements of water column optical nator, was optimal for both linear and quadratic formula-
properties, together with visual inspection of WV2 images, tions. The simple d vs. X regression yielded a strong
implied that flow depths could be inferred from satellite (R2 = 0.827) linear relationship between depth and the band
data. In this section, we report depth retrieval results from ratio, but adding an X2 term only slightly increased predic-
the Snake River and SBC, with a focus on accuracy assess- tive power (R2 = 0.864). Closer inspection of the OBRA
ment of depth estimates obtained using various calibration matrices indicated similar spectral patterns, but the NIR
approaches and image data types. bands became more useful when the X2 term was included.
3.2.1. Snake River To an extent, allowing for a non-linear d vs. X relation
3.2.1.1. Depth Retrieval Calibration Approaches accounted for saturation of the radiance signal in deeper
[36] The most direct means of mapping bathymetry from water in the NIR. Nevertheless, the marginal improvement
remotely sensed data involved extracting image spectra from provided by the quadratic term implied that the additional
the locations of field-based depth measurements and using complexity of this approach was not justified.
these data to calibrate a relationship between d and the [37] Bathymetric maps generated from the WV2 satellite
image-derived quantity X defined by equation (3). The image of Rusty Bend using linear and quadratic OBRA

12 of 24
F04024 LEGLEITER AND OVERSTREET: MAPPING RIVER BATHYMETRY FROM SPACE F04024

relations are shown in Figures 10a and 10b, respectively.


The maps were very similar to one another, with the linear
formulation yielding slightly shallower depth estimates in
the pool along the outer bank and over the bar surface on the
right (north) side of the channel, relative to the quadratic
OBRA. Both techniques produced realistic depictions of the
gross morphology, with an asymmetric cross-sectional shape
extending from above the apex through the middle of the
bend before the thalweg shifted toward the right at the lower
end of the reach. Depth retrieval accuracy was assessed via
comparison with pixel-scale mean depths obtained by OBK
of the field survey data. The resulting residual maps are
shown in Figures 11a and 11b, where negative residuals
(bright red tones) represent over-predictions of depth and
positive residuals (darker blue tones) represent under-
predictions of depth. Again, spatial patterns for the two
versions of OBRA were similar, with the linear d vs. X
relation leading to more extensive underestimates of depth
along the outer bank in the upper end of the reach and a
greater number of overestimates along the left bank past the
bend apex. Most depth retrieval residuals were on the order
of 0.25 m but locally ranged as high as 0.6 m, nearly half the
mean depth of 1.23 m, in the thalweg above the apex and on
the outer bank through the lower portion of Rusty Bend.
[38] For the most part, depth estimates from both linear
and quadratic OBRA were reliable, as indicated by high R2
values for observed vs. predicted (OP) regressions based on
field measurements set aside for validation (Figure 12 and
Table 2). Moreover, small mean errors and OP intercept and
slope values near 0 and 1, respectively, implied that depth
estimates were unbiased on average. The OP regression plot
in Figure 12a revealed a curved trend for linear OBRA,
however, with systematic under-prediction of depth in both
shallow and deep water. This trend arose from the non-lin-
earity of the relationship between d and X and was effec-
tively removed by including an X2 term in the OBRA
regression (Figure 12b). Nevertheless, the depth retrieval
RMSE and OP regression standard errors were only slightly
less for the quadratic vs. linear OBRA, and a typical error of
0.22 m would be less than 20% of the reach-averaged mean
depth. Because the quadratic formulation did not signifi-
cantly improve accuracy, the standard linear OBRA appeared
to be well-suited for bathymetric mapping in larger, deeper
gravel bed rivers such as the Snake, although an X2 term
could prove useful in streams with greater depths.
[39] An alternative strategy for remote mapping of river
bathymetry would involve making direct measurements of
depth and reflectance and using field spectra to develop d
versus X relations that could then be applied to remotely
sensed data. If the spectral response function of the sensor
were known, the field spectra could be convolved to match
the instrument’s band passes, as shown in Figures 3a and 3b
for the WV2 satellite. Ideally, OBRA of the convolved field
spectra would result in a calibrated relationship between d
and X that could be applied directly to images of the river

Figure 9. Optimal band ratio analysis (OBRA) from Rusty


Bend of the Snake River: (a) field spectra convolved to
match the band passes of the WorldView-2 (WV2) sensor;
(b) linear OBRA of WV2 satellite image spectra; and
(c) quadratic OBRA of WV2 satellite image spectra.

13 of 24
F04024 LEGLEITER AND OVERSTREET: MAPPING RIVER BATHYMETRY FROM SPACE F04024

Figure 10. Image-derived depth maps for Rusty Bend of the Snake River produced using various calibra-
tion approaches and image data types. All maps have a common color scale shown at the bottom. Flow is
from right to left.

from which the field spectra were collected and potentially outer bank was not well-resolved by depth estimates based
other streams as well. To assess the feasibility of this on convolved field spectra, and depths on the shallow bar
approach, we measured depth and reflectance on transects surface tended to be over-predicted. This pattern was evident
across Rusty Bend, convolved the spectra to the WV2 in the residual map in Figure 11c, which featured under-
bands, and performed OBRA. This analysis is summarized estimates on the order of 0.6 m in the pool above the bend
in Figure 9a, which indicated a strong linear relation apex and overestimates of similar magnitude at the lower end
(R2 = 0.839) between d and X, defined as the logarithm of of the reach. The OP regression for these depth estimates
the ratio of the green and red bands. This R2 value was nearly yielded a lower R2 value of 0.75, a large negative intercept
as high as that associated with the original field spectra term, and a slope much greater than 1 (Figure 12c), implying
(Figure 7a) and slightly better than that associated with the that predictions based on convolved field spectra were
linear OBRA of image spectra (Figure 9b). Applying the biased.
OBRA relation from Figure 9a to the WV2 image resulted in [40] In an effort to account for this bias, we adjusted the
the bathymetric map in Figure 10c, which has the same color image data to better match the field spectra as described in
scaling as the maps produced via OBRA of image spectra. Section 2.6 and illustrated in Figure 3. This adjustment,
A comparison of these maps indicated that the pool along the which could be considered a simple means of radiometric

14 of 24
F04024 LEGLEITER AND OVERSTREET: MAPPING RIVER BATHYMETRY FROM SPACE F04024

Figure 11. Maps of depth retrieval residuals, defined as the difference between the field-based depth
map and the image-derived depth estimates, for Rusty Bend of the Snake River produced using various
calibration approaches and image data types. All maps have a common color scale shown at the bottom.
Flow is from right to left.

calibration and atmospheric correction, served to modify the Accuracy assessment using the validation subset of the field
image so that the d vs. X relation derived from the convolved survey resulted in OP regression intercept and slope values
field spectra (Figure 9a) could be applied directly to the closer to 0 and 1, respectively, implying that the image
WV2 scene. The resulting bathymetric map, shown in correction was effective in removing the bias associated with
Figure 9d, featured greater depth estimates than did the calibration based on field spectra, but the OP R2 improved
original, uncorrected image but still failed to resolve the full only marginally, from 0.75 to 0.77. The RMSE and OP
depth of the pool along the outer bank. On the opposite side regression standard errors for both depth retrieval methods
of the channel, depths tended to be under-predicted, based on convolved field spectra were greater than those for
including some negative estimates. The spatial pattern of OBRA of image spectra; typical errors on the order of 0.3 m
these errors is illustrated in Figure 11d, which shows that would be approximately 25% of the mean depth for the
even after modifying the image data to better match the reach. These results imply that calibration via field spec-
convolved field spectra, pool depths were underestimated by troscopy was a plausible strategy, especially when the image
0.5 m or more throughout much of the bend; overestimates was corrected to better match the field spectra, but this
of 0.5 m or more occurred at the lower end of the reach. approach was less reliable than OBRA of image spectra.

15 of 24
F04024 LEGLEITER AND OVERSTREET: MAPPING RIVER BATHYMETRY FROM SPACE F04024

Figure 12. Depth retrieval validation for Rusty Bend of the Snake River for the various calibration
approaches and image data types labeled for each plot. Each plot represents the results of an observed
versus predicted (OP) regression as well as the one-to-one line of perfect agreement for comparison.

16 of 24
F04024 LEGLEITER AND OVERSTREET: MAPPING RIVER BATHYMETRY FROM SPACE F04024

Table 2. Depth Retrieval Accuracy Assessment for Various Calibration Methods and Image Types
Num.b Den.c Mean RMSE OP Std. OP OP Intercept
Reach Method OBRAaR2 (nm) (nm) Error (m) (m) OPdR2 Error (m) Slope (m)
Rusty Lineare 0.827 480 605 0.004 0.238 0.822 0.238 0.993 0.005
Quadratice 0.864 480 605 0.003 0.212 0.859 0.212 0.977 0.026
Fieldf 0.839 545 605 0.056 0.305 0.753 0.280 1.326 0.476
Field + imageg 0.839 545 605 0.058 0.271 0.769 0.271 1.005 0.052
Panh N/A N/A N/A 0.032 0.327 0.666 0.327 0.952 0.027
Pan-sharpi 0.794 480 605 0.009 0.254 0.801 0.254 0.992 0.001
Footbridge Lineare 0.585 545 725 0.001 0.094 0.594 0.110 0.987 0.004
Fieldf 0.962 660 725 0.026 0.195 0.452 0.110 0.380 0.157
Panh N/A N/A N/A 0.003 0.115 0.414 0.115 0.928 0.020
Pan sharpi 0.578 545 725 0.003 0.110 0.504 0.110 0.969 0.011
a
Optimal Band Ratio Analysis.
b
Numerator wavelength for optimal band ratio.
c
Denominator wavelength for optimal band ratio.
d
Observed versus predicted regression.
e
Linear and quadratic refer to linear and quadratic OBRA of image spectra.
f
Field refers to OBRA of field spectra convolved to match WorldView-2 sensor band passes.
g
Field + image is similar but refers to application of the OBRA relation from convolved field spectra to an image corrected to better match the field
spectra.
h
Pan refers to depth retrieval via the linear transform algorithm applied to the panchromatic image.
i
Pan-sharp refers to linear OBRA of image spectra from a pan-sharpened multispectral image.

3.2.1.2. Image Data Types derived depths indicated a weaker, but still fairly strong
[41] Although depth retrieval via OBRA requires multiple agreement between observations and predictions (R2 = 0.67),
spectral bands, bathymetric maps also can be generated from and the OP regression equation did not imply any systematic
single-band or panchromatic gray scale images using the bias. Closer examination of Figure 12e revealed that this
linear transform algorithm introduced by Lyzenga [1981]. agreement deteriorated considerably in deeper water, with
In the context of satellite imagery, this capability is poten- the image-derived estimates failing to increase as rapidly as
tially significant because many spaceborne sensors feature a the field-surveyed depths for d > 1 m.
panchromatic band with a greater spatial resolution than the [43] The bathymetric map derived from the pan-sharpened
individual multispectral bands; for the WV2 system, the image (Figure 10f) included a broader area of deep water
pixel sizes for the panchromatic and multispectral images are along the outer bank than did the map produced from the
0.5 m and 2 m, respectively. If spectral information is not panchromatic image, suggesting that the additional spectral
essential for accurate depth retrieval, application of the linear information resulted in a more accurate representation of the
transform (equation (4)) might enable more effective map- thalweg. The grainy texture of the panchromatic image
ping of smaller streams and enhanced spatial detail in larger persisted in the pan-sharpened data, however, and appeared
rivers. If spectral information does prove critical, a hybrid to be even more pronounced over the point bar on the right
approach based on a pan-sharpened image that features the side of the channel at the lower end of the reach. The
high spatial resolution of the panchromatic image but also residual map in Figure 11f was similar to that associated
incorporates the multispectral data could prove to be most with the panchromatic image, but the large underestimates in
useful. the upper portion of the bend were less extensive. Accuracy
[42] We explored this possibility and evaluated the rela- assessment of depth estimates from the pan-sharpened mul-
tive significance of spectral and spatial resolution by pro- tispectral image, which had a pixel edge dimension four
ducing bathymetric maps from 0.5 m-pixel panchromatic times smaller than the original multispectral data, yielded an
and pan-sharpened WV2 images of the Snake River. These OP regression R2 of 0.80 (Figure 12f), which was a signifi-
maps are included in Figures 10e and 10f, along with the cant improvement over the panchromatic image but was less
bathymetry inferred from the original, 2 m-pixel multispec- than that associated with the 2 m-pixel multispectral image.
tral images. For the panchromatic scene, depth retrieval via This result implied that greater spectral information content
the linear transform algorithm captured the general mor- was more important than enhanced spatial detail for depth
phology of the reach but the map had a grainy appearance retrieval from this relatively large gravel bed river. The noise
due to high-frequency noise that persisted despite the introduced by fusing the panchromatic image with the mul-
application of a smoothing filter. The area of deeper flow tispectral data also might have contributed to the greater
along the outer bank was narrower in the bathymetric map depth retrieval errors associated with the pan-sharpened
derived from the panchromatic data than in the map based on image. In any case, our analysis suggested that the additional
the original multispectral image (Figure 10a), suggesting image processing required to generate the pan-sharpened
that spectral information was needed to resolve greater image was not justified and that depth retrieval from the
depths. The corresponding residual map (Figure 11e) high- original multispectral images might be both simpler and
lighted this pattern, with depth underestimates on the order more accurate in this setting.
of 0.6 m extending across a greater fraction of the channel 3.2.2. Soda Butte Creek
width over the upper half of the reach. Past the bend apex, 3.2.2.1. Depth Retrieval Calibration Approaches
depths were over-predicted in a large area along the left side [44] The Footbridge Reach presented a different type of
of the channel. Accuracy assessment of the linear transform- challenge for remote sensing of river bathymetry: for this

17 of 24
F04024 LEGLEITER AND OVERSTREET: MAPPING RIVER BATHYMETRY FROM SPACE F04024

Figure 13. Optimal band ratio analysis (OBRA) from the Footbridge Reach of Soda Butte Creek:
(a) OBRA of image spectra; and (b) field spectra convolved to match the WorldView-2 sensor bands.

small gravel bed stream, the pixel size was a significant the two Footbridge Reach data sets suggested that bathy-
fraction of the mean channel width, particularly under the metric mapping based on field spectra might be more
low-flow conditions when the WV2 image was acquired. effective than extracting spectra directly from an image.
Moreover, at this discharge, SBC split into two separate Whereas many of the image spectra were from mixed pixels
channels, each of which was further divided by mid-channel contaminated by radiance from adjacent terrestrial features,
bars. As a result, the flow was quite shallow (Table 1) and each of the field spectra sampled a smaller area entirely
deriving information on channel form thus required an within the wetted channel. Atmospheric effects present in
imaging system capable of resolving subtle variations in the satellite image data but absent from the field measure-
depth. Based on the results for the Snake River, and because ments also might have contributed to the inferior OBRA
SBC was so much shallower, we focused on the standard results for the image spectra.
linear formulation of the OBRA algorithm and did not pur- [46] Bathymetric maps produced using these two calibra-
sue the quadratic alternative. Similarly, because our analysis tion methods are presented in Figures 14a and 14b, both of
of convolved field spectra from the Snake River indicated which have a common color scale with an upper limit set to
that depth estimates would be biased unless the image data the maximum depth observed in the field. In general, both
were adjusted to match the field spectra, we considered only maps captured the gross morphology of the reach, with
the latter approach for SBC. deeper flow along the left (east) bank at the upper end of our
[45] OBRA results obtained using spectra extracted from study area, in a pool on the far right channel near the
the WV2 image or measured directly in the field are shown entrance to the bend, in the lower half of the left chute
in Figures 13a and 13b, respectively. For the image spectra, channel, and at the bottom of the reach where the channels
the highest d vs. X regression R2 occurred for a green converge and the stream enters a curve to the right. Aside
numerator and NIR numerator (the sensor’s ‘red edge’ band) from these areas, depths were very shallow, on the order of
but was only 0.585, significantly less than the 0.827 R2 on 0.2 m. Although both OBRA of image spectra and applica-
the Snake River. The OBRA matrix in Figure 13a indicated tion of a d vs. X calibration relation derived from field
that this NIR band was by far the most useful denominator spectra to the corrected WV2 image produced reasonable
wavelength with any visible numerator; other denominator spatial patterns, the absolute magnitudes of the depth esti-
bands produced much weaker d vs. X relations. For the mates were less reliable, especially for calibration based on
convolved field spectra, the OBRA results were more field spectra. The bathymetric map shown in Figure 14b
encouraging, with an optimal R2 value of 0.962 for a red exceeded the specified color limits at both the shallow and
numerator band and the same 725 nm denominator as for the deep ends of the spectrum, resulting in the dark red tones
image spectra. Unlike the image spectra, however, a broader most notable over the point bar near the apex and the dark
range of bands produced strong, linear relationships between blue tones at the upper and lower ends of the reach. Depths
d and X, including longer NIR wavelengths. This result was estimated from image spectra, in contrast, fell within the
expected because strong absorption by pure water in the NIR range of observed depths and were more accurate.
caused small changes in depth to be expressed as relatively [47] Differences between the two approaches also were
large changes in reflectance; because depths in this reach highlighted by the residual maps shown in Figures 15a and
were so shallow, saturation of the NIR reflectance signal was 15b. For the image-based algorithm, most of the residuals
less of an issue than along the deeper Snake River. OBRA of were near zero, with a tendency to over-predict depth in

18 of 24
F04024 LEGLEITER AND OVERSTREET: MAPPING RIVER BATHYMETRY FROM SPACE F04024

Figure 14. Image-derived depth maps for the Footbridge Reach of Soda Butte Creek produced using
various calibration approaches and image data types. All maps have a common color scale shown at
the bottom. Flow is from top to bottom.

shallow areas and under-predict in pools. When the OBRA limiting factor for this small stream. Similarly, the inability
relation from the field spectra was applied to the corrected of either method to detect the deep pool in the middle of the
image, the overestimates of depth in shallow riffles and bars left chute channel, evident as a dark blue area on the residual
were more pronounced, as was the under-prediction in the maps, also implied that the 2 m pixel size might not have
left chute channel and along the outer banks at both the been adequate for mapping SBC under low-flow conditions.
upper and lower ends of the reach. Large positive residuals [48] The accuracy assessment summarized in Figures 16a
near steep cutbanks suggested that the presence of both and 16b yielded further insight regarding discrepancies
bright terrestrial features and deep water within mixed pixels between bathymetric maps produced from field vs. image
might have lead to unreliable depth estimates along channel spectra. Regression of observed vs. predicted depths yielded
margins and that sensor spatial resolution might have been a a stronger correlation (R2 = 0.59) for depth retrieval based on

Figure 15. Maps of depth retrieval residuals, defined as the difference between the field-based depth
map and the image-derived depth estimates, for the Footbridge Reach of Soda Butte Creek produced using
various calibration approaches and image data types. All maps have a common color scale shown at the
bottom. Flow is from top to bottom.

19 of 24
F04024 LEGLEITER AND OVERSTREET: MAPPING RIVER BATHYMETRY FROM SPACE F04024

Figure 16. Depth retrieval validation for the Footbridge Reach of Soda Butte Creek for the various cal-
ibration approaches and image data types labeled for each panel. Each plot represents the results of an
observed versus predicted (OP) regression as well as the one-to-one line of perfect agreement for compar-
ison. Note that Figure 16b has different axis limits than the other plots.

image spectra than when using the convolved field spectra use of a near-infrared band that might have been more
for calibration, which resulted in an R2 of only 0.45. More strongly affected by residual atmospheric effects not fully
importantly, however, the image-based estimates were accounted for in the reflectance retrieval algorithm used by
unbiased, with an OP regression intercept near 0 and slope DigitalGlobe.
near 1. In contrast, depth estimates derived using the cali- 3.2.2.2. Image Data Types
bration relation for the convolved field spectra resulted in a [49] Given the small size of SBC and the modest depth
positive intercept term nearly as large as the mean depth for retrieval performance of the 2 m-pixel WV2 data, we eval-
the reach and a much smaller slope, implying biased esti- uated whether images with greater spatial resolution might
mates. This bias lead to over-predictions of depth in shallow prove more useful for bathymetric mapping. Depth maps
areas and under-predictions in deep areas and persisted produced from 0.5 m-pixel panchromatic and pan-sharpened
despite the correction applied to the image to force closer images are shown in Figures 14c and 14d, with the same
agreement with the field spectra. Although the simple image color scaling as used for the maps generated from the
adjustment outlined in Section 2.6 allowed for reasonably coarser-resolution data. Overall spatial patterns were similar,
accurate depth retrieval for the Snake River, this technique but the smaller pixel size revealed some details of the mor-
was not effective on SBC, implying that more sophisticated phology that were not evident in Figures 14a and 14b. For
radiometric calibration and atmospheric correction might example, the depth map generated by applying the linear
be required before relationships derived from field spectra transform to the panchromatic image captured shoaling of
can be applied to remotely sensed images. Another factor the flow onto the point bar in the middle of the three chan-
that might have contributed to biased estimates was the nels present at the bend apex, a feature that was not resolved

20 of 24
F04024 LEGLEITER AND OVERSTREET: MAPPING RIVER BATHYMETRY FROM SPACE F04024

by any of the other image types. The linear transform failed band ratio-based algorithm, calibrated using either spectra
to detect the full depth of pools located in the left chute extracted from the image or measured directly in the
channel and at the lower end of the reach, however. The pan- field, provided reliable depth estimates. The results of this
sharpened multispectral image did not provide quite as much study thus imply that satellite-based mapping of river
detail but yielded greater depth estimates in the pools than bathymetry could become a viable tool for river research
did the panchromatic data. The residual maps in Figure 15 and management.
indicated that depth retrieval errors from the higher resolu-
tion images were of a similar magnitude as those associated 4.1. Constraints on Spectrally-Based Bathymetric
with the original multispectral image. Over-predictions of Mapping
depth in the riffle at the upper end of the reach and [51] Although the potential for remote sensing methods to
approaching the bar at the bend apex and under-predictions contribute to the riverine sciences is justifiable cause for
of depth throughout the lower portion of the reach were excitement [Marcus and Fonstad, 2010], we advocate a
salient in the residual maps for both the panchromatic and cautious approach that acknowledges the constraints as well
pan-sharpened images. The accuracy assessment summa- as the capabilities associated with this new technology. The
rized in Figures 16c and 16d indicated that depth estimates retrieval of water depth from passive optical image data, for
were unbiased in both cases but were more reliable for the example, is subject to some important caveats. The most
pan-sharpened than for the panchromatic image. OP regres- significant constraint on the spectrally-based approach is the
sion R2 values of 0.50 and 0.41, respectively, were signifi- limited range of stream conditions under which bathymetry
cantly less than the 0.59 obtained using image spectra from can be mapped reliably. For rivers having greater depths (on
the original multispectral scene. Typical errors of 0.11 m the order of several meters) and/or more turbid water due to
were half the mean flow depth for the reach. These results higher amounts of suspended sediment and/or organic mat-
imply that even in a smaller stream such as SBC, the addi- ter, accurate depth estimates are less likely, and much of the
tional spatial detail provided by the panchromatic image did channel might exceed the maximum detectable depth [e.g.,
not translate into improved depth retrieval performance. Legleiter et al., 2011b, 2011a]. Overhanging vegetation,
Instead, spectral information was essential for reliable shadows, mixed pixels along channel margins, dark sub-
bathymetric mapping. strates (resulting in low LB(l) values), sun glint from the
water surface, and hazy atmospheric conditions can also
compromise, if not preclude, effective depth retrieval
4. Discussion [Legleiter et al., 2009].
[50] Efforts to characterize the complex interactions [52] In addition to these environmental factors, the extent
among flow, morphology, and sediment transport that shape to which an image data set can satisfy the information
alluvial river channels have often been compromised by the requirements of a particular investigation also depends on
difficulty of acquiring basic measurements of channel form. the sensor itself. For example, the imaging system must have
As a result, field studies typically have examined only short sufficient spatial resolution to detect the channel features of
reaches in isolation. Important, increasingly interdisciplin- interest. Similarly, instruments with greater spectral resolu-
ary research opportunities thus exist at larger segment and tion might enable more accurate bathymetric mapping by
watershed scales [e.g., Fausch et al., 2002; Carbonneau providing radiance measurements in a larger number of
et al., 2011]. For example, geomorphologists might seek narrower wavelength bands, some of which could be highly
to examine how the detailed process mechanics documented responsive to changes in depth. Radiometric resolution
through reach-scale studies self-organize within a catchment refers to a sensor’s ability to detect subtle differences in
to create emergent patterns in channels and landscapes that radiance, such as those associated with small variations in
have tended to be described only in conceptual or theoretical water depth, and thus exerts a primary control on both the
terms [e.g., Church, 2002; Benda et al., 2004]. Similarly, precision of depth estimates and the maximum detectable
greater insight on the distribution of in-stream habitat within depth. Forward image modeling, which involves simulating
a watershed would help biologists to understand the move- an image “from the streambed up” given information on the
ment of fish throughout their life histories [e.g., Ganio et al., morphology and optical characteristics of the channel and
2005]. In an applied context, a more synoptic perspective on the technical specifications of the sensor, provides a means
fluvial systems would help to provide a holistic, integrated of determining a priori the accuracy, precision, and dynamic
context for the planning, implementation, and monitoring of range of image-derived depth estimates. This approach
river restoration projects [e.g., Beechie et al., 2010; Downs allows tradeoffs among spatial, spectral, and radiometric
et al., 2011]. For all of these purposes, the lack of an efficient resolution to be evaluated in the context of a specific river of
means of mapping river morphology has impeded progress interest [Legleiter and Roberts, 2009].
and the development of new techniques could thus stimulate [53] In this study, rather than resort to modeling, we made
significant advances in each of these fields [Fonstad and direct measurements of water column attenuation in a pair of
Marcus, 2010]. Motivated by this prospect, this investiga- gravel bed rivers and used these data to quantify the limita-
tion demonstrated the feasibility of mapping the bathymetry tions of spectrally-based depth retrieval. Vertical profiles of
of clear-flowing gravel bed rivers from satellite image data. downwelling spectral irradiance were used to calculate
Using field measurements and WV2 imagery from two values of the diffuse attenuation coefficient Kd(l) that were
streams in the northern Rocky Mountains, we showed that in turn used to determine the smallest detectable change in
information on flow depth can be retrieved from multispec- depth and maximum detectable depth via the theoretical
tral data across a range of channel sizes from approximately framework established by Philpot [1989]; this analysis was
20–60 m in width and 0.2–1.25 m in mean depth. A simple performed for hypothetical imaging systems with specified

21 of 24
F04024 LEGLEITER AND OVERSTREET: MAPPING RIVER BATHYMETRY FROM SPACE F04024

levels of radiometric resolution. Our results indicated that for water strongly absorbs the NIR laser pulses emitted by these
these clear-flowing streams, a typical instrument with a instruments. To obtain a complete, hybrid representation
DLN(l)/LB(l) ratio of 0.01 would yield depth estimates with of the fluvial environment, LiDAR topography from
a precision on the order of 0.01–0.04 m and a maximum exposed bars and floodplains can be combined with
detectable depth ranging from 5 m in the green portion of the channel bathymetry derived from passive optical image
visible spectrum to 1.3 m in the NIR. More sensitive data, although such data fusion involves various technical
instrumentation would yield more precise estimates and challenges [Legleiter, 2012a]. Newly developed, water-
increase the dynamic range of depth retrieval, but, con- penetrating green LiDAR systems provide a more direct
versely, less sophisticated sensors would provide estimates solution for measuring both subaerial and submerged sur-
subject to greater uncertainty over a more restricted range of faces [e.g., McKean et al., 2008]. Originally developed for
depths. These theoretical calculations were consistent with coastal applications, these bathymetric LiDARs provide a
our analysis of actual satellite image data and corresponding relatively large laser spot size and low point density and
field-based depth measurements. Typical errors of 0.238 and tend to over-estimate bed elevations in riverine settings
0.094 m for the larger and smaller of our study streams, [Kinzel et al., 2007]. In addition, laser returns from the
respectively, corresponded to 20% and 42% of the mean water surface, water column, and streambed can be diffi-
flow depths for the two reaches examined in detail. Depth cult to distinguish in shallow channels [Kinzel et al., 2012].
estimates calibrated using field measurements representative Thus, although green LiDAR has outstanding potential for
of the overall distribution of depths within each reach were measuring riverine topography, these systems remain
less reliable in deeper water, so the maximum detectable experimental and have yet to achieve operational status as a
depth was not as well-constrained by our data, but pools viable monitoring tool.
over 2.5 m deep were captured by satellite-based bathy- [56] Passive optical remote sensing, in contrast, has
metric maps. A combination of in situ observations of water become routine, with image data acquired on a regular basis
column optical properties, theoretical calculations, and and made freely available to the public. For example, aerial
careful accuracy assessment will be most effective in defin- photography acquired through the National Agriculture
ing the constraints associated with remote sensing of river Imagery Program is accessible online and has been shown to
bathymetry. provide reasonably accurate bathymetric information for
clear-flowing gravel bed rivers. The low radiometric reso-
4.2. Alternative Approaches to Remote Measurement lution of these basic image data resulted in saturation of the
of River Morphology radiance signal, however, and the full depth of pools could
[54] This study focused on retrieving water depth from not be detected [Legleiter, 2012c]. For focused scientific
passive optical image data using a single, relatively simple investigations or any application for which precise estimates
band ratio algorithm, but several alternative strategies for across a broad range of depths are necessary, acquiring task-
remote sensing of river morphology are available as well. specific data with more advanced multi or hyperspectral
For example, the same type of forward image modeling sensors might be more appropriate. This type of airborne
described in Section 4.1 forms the basis of more advanced data collection requires careful planning and can be com-
spectrum-matching methods now favored by the coastal plicated by a number of different factors. In our experience,
research community for mapping depth, bottom composi- logistical coordination of pilots, planes, and instrumentation
tion, and concentrations of various optically significant can prove difficult, particularly if image acquisition is to be
constituents of the water column [e.g., Lee et al., 1999; synchronized with field-based measurements for purposes of
Mobley et al., 2005]. In theory, these techniques could be calibration and validation. Weather, mechanical problems,
adapted for application to riverine environments, but this and other unforeseen circumstances can derail even the most
prospect has yet to be explored. Implementing such an well-thought out missions. In sensitive areas, such as the
approach would require additional data on the spectral National Parks where we conducted this investigation, flight
characteristics of different fluvial substrates and water types, permits must be secured and restrictions on flying height
as well as significant modeling effort. In this study, we made above terrain can compromise spatial resolution and sensor
some initial progress toward this goal by making field signal-to-noise. For larger study areas or channels that do not
measurements of reflectance and apparent and inherent follow a straight course, multiple flight lines might be
optical properties of the water column for a pair of clear- needed, resulting in multiple images that must be geo-
flowing gravel bed rivers. If a suitable database can be referenced and mosaicked and might not be consistent with
developed, this physics-based approach could allow for one another in terms of atmospheric conditions and viewing
greater generality and thus eliminate the need to coordinate and illumination geometry. For these reasons, airborne data
remotely sensed data collection with field measurements to might not be optimal for remote sensing of rivers.
establish relationships between image-derived quantities and
observed flow depths. Continued reliance upon this type of 4.3. Potential Advantages of Mapping River
empirical calibration, which involves regressing in situ depth Bathymetry From Space
measurements against image pixel values, would undermine [57] Satellite imagery could prove to be a viable alternative
one of the principal advantages of remote sensing. with several distinct advantages. A number of commercial
[55] A complementary technology for characterizing river satellites, such as QuickBird, GeoEye, and WorldView, now
morphology is light detection and ranging, or LiDAR. provide the kind of spatial resolution required to map small-
Although LiDAR has become a preferred method of mea- to medium-sized gravel bed rivers, with pixel sizes of 2 m or
suring topography [Slatton et al., 2007], typical LiDAR less for multispectral images and 0.5 m for panchromatic
systems cannot measure submerged bed elevations because images. These sensors also feature off-nadir pointing

22 of 24
F04024 LEGLEITER AND OVERSTREET: MAPPING RIVER BATHYMETRY FROM SPACE F04024

capabilities that provide greater flexibility for tasked data principal conclusion of this investigation is that, under
collection during an acquisition window specified by the appropriate circumstances that include clear water condi-
user. Because the platform is already in orbit, logistical pro- tions and shallow to moderate depths, river bathymetry can
blems associated with remotely sensed data collection are be mapped from space, provided that field measurements of
less likely and obtaining flight permits is not an issue; plan- depth are available for calibration. By exploiting this capa-
ning can instead focus on field activities. Satellite images bility, while also acknowledging the limitations associated
typically encompass a larger area than could be acquired with remote sensing, scientists might gain novel insight on
along an aerial flight line, so longer river segments or even the dynamics of certain types of fluvial systems.
entire watersheds can be captured in a single scene with
uniform radiometry, rather than a mosaic of multiple flight [59] Acknowledgments. This investigation was supported by a grant
from the Office of Naval Research Littoral Geosciences and Optics Program
strips. For cases where the entire image is not relevant, (grant N000141010873). The National Park Service granted permission to
commercial providers often market data on a per km2 basis to conduct research in Yellowstone and Grand Teton National Parks. The Uni-
match the user’s area of interest, so satellite imagery might be versity of Wyoming-National Park Service Research Center and the Yel-
lowstone Ecological Research Center provided logistical support. The ac-
more cost-effective as well. Because these data products 9 instrument was borrowed from the Naval Research Laboratory and the
often are geo-referenced and atmospherically corrected, sat- United States Geological Survey loaned suspended sediment sampling
ellite images can be used as delivered, without the need for equipment and processed water samples. Gregory Miecznik of DigitalGlobe
coordinated acquisition of the WorldView-2 images used in this study. C.L.
extensive pre-processing that could require specialized soft- Rawlins and Floyd Legleiter assisted with field work. The editor, associate
ware and expertise. In addition, obtaining repeat coverage to editor, and three anonymous reviewers provided useful comments on an
characterize channel change is more feasible because orbiting earlier version of this paper.
satellites can be tasked on an as-needed basis in response to a
geomorphically significant event, whereas acquiring post-
flood aerial data would involve a second round of complex References
logistical arrangements. For application-oriented users more Ashmore, P. E., and M. Church (1998), Sediment transport and river mor-
concerned with information derived via remote sensing and phology: A paradigm for study, in Gravel-Bed Rivers in the Environment,
less interested in the details of flight planning and image edited by P. C. Klingeman et al., pp. 115–140, Water Resour. Publ.,
Highlands Ranch, Colo.
processing, satellite data might provide a simpler, more Atkinson, P. M., and P. J. Curran (1995), Defining an optimal size of sup-
consistent solution. port for remote sensing investigations, IEEE Trans. Geosci. Remote
Sens., 33(3), 768–776.
Bailly, J.-S., Y. L. Coarer, P. Languille, C.-J. Stigermark, and T. Allouis
5. Conclusion (2010), Geostatistical estimations of bathymetric LiDAR errors on rivers,
Earth Surf. Processes Landforms, 35(10), 1199–1210.
[58] Efforts to characterize and understand the morphol- Beechie, T. J., D. A. Sear, J. D. Olden, G. R. Pess, J. M. Buffington, H.
ogy and dynamics of alluvial river channels are often com- Moir, P. Roni, and M. M. Pollock (2010), Process-based principles for
restoring river ecosystems, BioScience, 60(3), 209–222.
promised by the difficulty of measuring their form and Benda, L., N. L. Poff, D. Miller, T. Dunne, G. Reeves, G. Pess, and M.
behavior via conventional, ground-based field methods. This Pollock (2004), The network dynamics hypothesis: How channel net-
study explored the potential to map river bathymetry from works structure riverine habitats, Bioscience, 54(5), 413–427.
Brasington, J., B. T. Rumsby, and R. A. McVey (2000), Monitoring and
passive optical images acquired from spaceborne satellite modelling morphological change in a braided gravel-bed river using high
platforms that provide high resolution data, such as the 2 m- resolution GPS-based survey, Earth Surf. Processes Landforms, 25(9),
pixel WorldView2 images evaluated here. Our results indi- 973–990.
cate that water depths in clear-flowing, mid- to large-sized Carbonneau, P., M. A. Fonstad, W. A. Marcus, and S. J. Dugdale (2011),
Making riverscapes real, Geomorphology, 137(1), 74–86.
gravel bed rivers with depths <3 m and widths <60 m can be Church, M. (2002), Geomorphic thresholds in riverine landscapes, Fresh-
estimated reliably from multispectral data. Direct measure- water Biol., 47(4), 541–557.
ments of water column optical properties were used to Dierssen, H. M., R. C. Zimmerman, R. A. Leathers, T. V. Downes, and
C. O. Davis (2003), Ocean color remote sensing of seagrass and
quantify some of the key constraints associated with such bathymetry in the Bahamas Banks by high-resolution airborne imagery,
spectrally-based depth retrieval. For typical levels of sensor Limnol. Oceanogr., 48(1), 444–455.
radiometric resolution, the smallest detectable change in Downs, P. W., M. S. Singer, B. K. Orr, Z. E. Diggory, and T. C. Church
depth was calculated to be on the order of 0.01–0.04 m and (2011), Restoring ecological integrity in highly regulated rivers: The
role of baseline data and analytical references, Environ. Manage., 48(4),
the maximum detectable depth to vary with wavelength from 847–864.
>5 m for green bands to <2 m in the NIR. A simple, band Erwin, S. O., J. C. Schmidt, and N. C. Nelson (2011), Downstream effects
ratio-based algorithm for selecting appropriate combinations of impounding a natural lake: The Snake River downstream from Jackson
Lake Dam, Wyoming, USA, Earth Surf. Processes Landforms, 36(11),
of wavelengths and calibrating relationships between field 1421–1434.
measurements of depth and image-derived quantities was Fausch, K. D., C. E. Torgersen, C. V. Baxter, and H. W. Li (2002), Land-
shown to be effective when applied to spectra extracted from scapes to riverscapes: Bridging the gap between research and conserva-
tion of stream fishes, Bioscience, 52(6), 483–498.
an image or recorded directly in the field, although a dif- Flener, C., E. Lotsari, P. Alho, and J. Kayhko (2012), Comparison of empir-
ferent pair of bands was selected for each of the streams ical and theoretical remote sensing based bathymetry models in river
examined. Adding a quadratic term to these relationships environments, River Res. Appl., 28(1), 118–133.
provided only a marginal improvement in bathymetric Fonstad, M. A., and W. A. Marcus (2005), Remote sensing of stream depths
with hydraulically assisted bathymetry (HAB) models, Geomorphology,
accuracy in a deeper river. Similarly, neither panchromatic 72(1–4), 320–339.
nor pan-sharpened multispectral images with greater spatial Fonstad, M. A., and W. A. Marcus (2010), High resolution, basin extent
resolution yielded more accurate depth estimates, even in a observations and implications for understanding river form and process,
Earth Surf. Processes Landforms, 35(6), 680–698.
smaller stream for which the pixel size was 10% of the mean Ganio, L. M., C. E. Torgersen, and R. E. Gresswell (2005), A geostatistical
channel width. These results implied that spectral informa- approach for describing spatial pattern in stream networks, Front. Ecol.
tion content was crucial to reliable depth retrieval. The Environ., 3(3), 138–144.

23 of 24
F04024 LEGLEITER AND OVERSTREET: MAPPING RIVER BATHYMETRY FROM SPACE F04024

Goovaerts, P. (1997), Geostatistics for Natural Resources Evaluation, bathymetry and topography from an unmanned mapping controlled plat-
Oxford Univ. Press, New York. form, Earth Surf. Processes Landforms, 32(11), 1705–1725.
Ham, D. G., and M. Church (2000), Bed-material transport estimated from Little, L. S., D. Edwards, and D. E. Porter (1997), Kriging in estuaries:
channel morphodynamics: Chilliwack River, British Columbia, Earth As the crow flies, or as the fish swims?, J. Exp. Mar. Biol. Ecol.,
Surf. Processes Landforms, 25(10), 1123–1142. 213(1), 1–11.
Hodge, R., J. Brasington, and K. Richards (2009), In situ characterization of Lyzenga, D. R. (1981), Remote sensing of bottom reflectance and water
grain-scale fluvial morphology using terrestrial laser scanning, Earth attenuation parameters in shallow water using aircraft and Landsat data,
Surf. Processes Landforms, 34(7), 954–968. Int. J. Remote Sens., 2(1), 71–82.
Keim, R. F., A. E. Skaugset, and D. S. Bateman (1999), Digital terrain mod- Marcus, W. A., and M. A. Fonstad (2008), Optical remote mapping of riv-
eling of small stream channels with a total-station theodolite, Adv. Water ers at sub-meter resolutions and watershed extents, Earth Surf. Processes
Resour., 23(1), 41–48. Landforms, 33(1), 4–24.
Kilham, N. E., D. Roberts, and M. B. Singer (2012), Remote sensing of sus- Marcus, W. A., and M. A. Fonstad (2010), Remote sensing of rivers: The
pended sediment concentration during turbid flood conditions on the emergence of a subdiscipline in the river sciences, Earth Surf. Processes
Feather River, California: A modeling approach, Water Resour. Res., Landforms, 35(15), 1867–1872.
48, W01521, doi: 10.1029/2011WR010391. Marcus, W. A., C. J. Legleiter, R. J. Aspinall, J. W. Boardman, and R. L.
Kinzel, P. J., C. W. Wright, J. M. Nelson, and A. R. Burman (2007), Eval- Crabtree (2003), High spatial resolution hyperspectral mapping of
uation of an experimental LiDAR for surveying a shallow, braided, sand- in-stream habitats, depths, and woody debris in mountain streams,
bedded river, J. Hydraul. Eng., 133(7), 838–842. Geomorphology, 55(1–4), 363–380.
Kinzel, P. J., C. J. Legleiter, and J. M. Nelson (2012), Mapping river Maritorena, S., A. Morel, and B. Gentili (1994), Diffuse-reflectance of oce-
bathymetry with a small footprint green LiDAR: Applications and chal- anic shallow waters: Influence of water depth and bottom albedo, Limnol.
lenges, J. Am. Water Resour. Assoc., doi:10.1111/jawr.12008, in press. Oceanogr., 39(7), 1689–1703.
Lane, S. N., J. H. Chandler, and K. S. Richards (1994), Developments in McKean, J. A., D. J. Isaak, and C. W. Wright (2008), Geomorphic controls
monitoring and modeling small-scale river bed topography, Earth Surf. on salmon nesting patterns described by a new, narrow-beam terrestrial-
Processes Landforms, 19(4), 349–368. aquatic LiDAR, Front. Ecol. Environ., 6(3), 125–130.
Lee, Z., K. L. Carder, C. D. Mobley, R. G. Steward, and J. S. Patch (1999), Mertes, L. A., M. O. Smith, and J. B. Adams (1993), Estimating suspended
Hyperspectral remote sensing for shallow waters: Deriving bottom depths sediment concentrations in surface waters of the Amazon River wetlands
and water properties by optimization, Appl. Opt., 38(18), 3831–3843. from Landsat images, Remote Sens. Environ., 43, 281–301.
Legleiter, C. J. (2012a), Remote measurement of river morphology via Meyer, G. A. (2001), Recent large-magnitude floods and their impact on
fusion of LiDAR topography and spectrally based bathymetry, Earth valley-floor environments of northeastern Yellowstone, Geomorphology,
Surf. Processes Landforms, 37(5), 499–518. 40, 271–290.
Legleiter, C. J. (2012b), A geostatistical framework for quantifying the Micheli, E. R., and J. W. Kirchner (2002), Effects of wet meadow riparian
reach-scale spatial structure of river morphology: 2. Application to vegetation on streambank erosion. 1. Remote sensing measurements of
restored and natural channels, Geomorphology, doi:10.1016/j.geomorph. streambank migration and erodibility, Earth Surf. Processes Landforms,
2012.01.017, in press. 27(6), 627–639.
Legleiter, C. J. (2012c), Mapping river depth from publicly available aerial Mishra, D. R., S. Narumalani, D. Rundquist, and M. Lawson (2005), Char-
images, River Res. Appl., doi:10.1002/rra.2560, in press. acterizing the vertical diffuse attenuation coefficient for downwelling
Legleiter, C. J. (2012d), A geostatistical framework for quantifying the reach- irradiance in coastal waters: Implications for water penetration by high
scale spatial structure of river morphology: 1. Variogram models, related resolution satellite data, ISPRS J. Photogramm. Remote Sens., 60(1),
metrics, and relation to channel form, Geomorphology, doi:10.1016/ 48–64.
j.geomorph.2012.01.016, in press. Mobley, C. D., and L. K. Sundman (2001), Hydrolight 4.2 User’s Guide, 2
Legleiter, C. J., and P. Kyriakidis (2006), Forward and inverse transforma- ed., Sequoia Sci., Redmond, Wash.
tions between Cartesian and channel-fitted coordinate systems for mean- Mobley, C. D., et al. (2005), Interpretation of hyperspectral remote-sensing
dering rivers, Math. Geol., 38(8), 927–958. imagery by spectrum matching and look-up tables, Appl. Opt., 44(17),
Legleiter, C. J., and P. C. Kyriakidis (2008), Spatial prediction of river chan- 3576–3592.
nel topography by kriging, Earth Surf. Processes Landforms, 33(6), Pardo-Iguzquiza, E. (1999), Varfit: A Fortran-77 program for fitting var-
841–867. iogram models by weighted least squares, Comput. Geosci., 25(3),
Legleiter, C. J., and D. A. Roberts (2005), Effects of channel morphology 251–261.
and sensor spatial resolution on image-derived depth estimates, Remote Philpot, W. D. (1989), Bathymetric mapping with passive multispectral
Sens. Environ., 95(2), 231–247. imagery, Appl. Opt., 28(8), 1569–1578.
Legleiter, C. J., and D. A. Roberts (2009), A forward image model for pas- Pineiro, G., S. Perelman, J. P. Guerschman, and J. M. Paruelo (2008), How
sive optical remote sensing of river bathymetry, Remote Sens. Environ., to evaluate models: Observed vs. predicted or predicted vs. observed?,
113(5), 1025–1045. Ecol. Modell., 216(3–4), 316–322.
Legleiter, C. J., D. A. Roberts, W. A. Marcus, and M. A. Fonstad (2004), Slatton, K. C., W. E. Carter, R. L. Shrestha, and W. Dietrich (2007), Air-
Passive optical remote sensing of river channel morphology and in- borne laser swath mapping: Achieving the resolution and accuracy
stream habitat: Physical basis and feasibility, Remote Sens. Environ., required for geosurficial research, Geophys. Res. Lett., 34, L23S10,
93(4), 493–510. doi:10.1029/2007GL031939.
Legleiter, C. J., D. A. Roberts, and R. L. Lawrence (2009), Spectrally based Smith, J. D., and S. R. McLean (1984), A model for flow in meandering
remote sensing of river bathymetry, Earth Surf. Processes Landforms, streams, Water Resour. Res., 20(9), 1301–1315.
34(8), 1039–1059. Wheaton, J. M., J. Brasington, S. E. Darby, and D. A. Sear (2010),
Legleiter, C. J., P. J. Kinzel, and B. T. Overstreet (2011a), Evaluating the Accounting for uncertainty in DEMs from repeat topographic surveys:
potential for remote bathymetric mapping of a turbid, sand-bed river: 2. Improved sediment budgets, Earth Surf. Processes Landforms, 35(2),
Application to hyperspectral image data from the Platte River, Water 136–156.
Resour. Res., 47, W09532, doi:10.1029/2011WR010592. Winterbottom, S. J., and D. J. Gilvear (1997), Quantification of channel bed
Legleiter, C. J., P. J. Kinzel, and B. T. Overstreet (2011b), Evaluating the morphology in gravel-bed rivers using airborne multispectral imagery
potential for remote bathymetric mapping of a turbid, sand-bed river: 1. and aerial photography, Regul. Rivers Res. Manage., 13(6), 489–499.
Field spectroscopy and radiative transfer modeling, Water Resour. Res., Zhang, X. F., J. C. H. Vaneijkeren, and A. W. Heemink (1995), On the
47, W09531, doi:10.1029/2011WR010591. weighted least-squares method for fitting a semivariogram model, Com-
Lejot, J., C. Delacourt, H. Piegay, T. Fournier, M.-L. Tremalo, and put. Geosci., 21(4), 605–608.
P. Allemand (2007), Very high spatial resolution imagery for channel

24 of 24

You might also like